Geophysical simulations in large-scale heterogeneous media are computationally expensive. Discrete models require large meshes leading to solve huge systems of equations.
Spectral element-agglomeration multigrid (SA-AMGe) is a discretization framework for diffusion equations that reduces significantly the computational cost while providing accurate solutions. SA-AMGe was developed at Lawrence Livermore National Laboratory as a two-grid method and implemented on a parallel distributed memory C++ library.
In collaboration with Berkeley Lab, the SA-AMGe library was optimized to improve its parallel performance and extended to a multilevel method. In this work, we demonstrate such multilevel features and analyze its parallel performance using a synthetic geophysical setup.
Note: Project and results were presented in a poster.